Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!tut.cis.ohio-state.edu!ucbvax!pasteur!ames!pacbell!hoptoad!peora!ucf-cs!farbmac!goldfarb
From: goldfarb@farbmac.UUCP (Benjamin I. Goldfarb)
Newsgroups: comp.unix.aux
Subject: Re: mkfs interleave factors
Summary: What does 2:1 interleave do to performance of MacOS partitions
	 on disks that have been reformatted?
Message-ID: <182@farbmac.UUCP>
Date: 26 May 89 01:47:19 GMT
References: <31426@apple.Apple.COM> <31611@apple.Apple.COM>
Distribution: usa
Organization: Longwood FL
Lines: 40

In article <31611@apple.Apple.COM>, dwells@Apple.COM (Dave Wells) writes:
> Our own HD 20 SC performs best at 2:1 with a mkfs gap of 1.  Format it with
> a 1:1 interleave and you'll nearly halve the performance.  Ouch.  Of course,
> our HD SC setup program almost forces a 1:1 interleave when 'initializing'
> on a MacII series.  You have to go through the cmd-I trick to use anything
> else.
> 
> Yep, customers have asked me for the optimal settings for this drive and
> A/UX.  I spent plenty of time playing around with whatever drives I could
> dig up - That's how I came up with 2:1 for this one.  Other low-performance
> drives will create the same problem when a 1:1 interleave is used.
> 
> -Dave

I have a Quantum Q280 internal 80 meg drive and a Quantum P80S external 80
meg drive.  I've formatted them both with 1:1 interleave and I use mkfs gap
of 1.  On one of the disks (the external), I have a 60 Meg MacOS partition
in addition to A/UX data.

Here's the question: if I accept Dave's advice and reformat both drives 
with 2:1 interleave, will I decrease performance when using MacOS?  I assume
the answer is yes or Apple wouldn't recommend 1:1 interleave for Mac IIs.
If that is the case, is what I have now a reasonable compromise?  Or would
it be better to leave the outboard drive as is and just reformat the internal?

Optimization of mkfs m and n parameters has been a black art ever since
mkfs was born.  In the ten years or so I've been associated with UNIX I've
never developed a firm grasp on this phase of performance tuning.  Certainly
the discussions on Usenet through the years have shown that the empirical
method still reigns supreme in determining m and n.  I doubt that things
in this area will change toward a pat solution anytime in the near future,
given the combinatorial possibilities with drive characteristics, file system 
organizations, processors, and the many other factors that influence file 
system performance.

------------------------------------------------------------------------------
Ben Goldfarb			uucp: {decvax,peora}!ucf-cs!farbmac!goldfarb
Department of Computer Science	Internet: goldfarb%farbmac.uucp@ucf-cs.ucf.edu
University of Central Florida	BITNET: GOLDFARB@UCF1VM
------------------------------------------------------------------------------